A Comparative Study of Different Feature Extractors for Parkinson's Disease Detection

For many decades, Parkinson's Disease (PD) has been a topic of discussion and research in the field of neurological disorders. It's one of the most common diseases associated with brain, but with no known cause and cure. There is no cure for the disease but with treatment of the symptoms during onset, progress of the disease can be controlled. But, an early-stage diagnosis of the disease is a must for proceeding with treatments to control the symptoms, making it a challenging task for the healthcare providers. There are many brain imaging tests like F-18 DOPA which can be helpful in diagnosing the disease, but early-stage detection can be more efficiently done using computer aided solutions. Most of these prominent solutions make use of AI based approaches which are trained over different kind of data. However, considering the image datasets there is an issue which is not yet addressed for the disease. There is a lack of comparative study on feature extraction techniques capable of handling scale and rotation invariance, the most important part for models running on image data. This work has addressed this issue by comparing performance of six different feature extractors: SIFT, ORB, KAZE, AKAZE, BRISK and FAST on a hand drawing dataset related to Parkinson's Disease.